The paper examines the duration of unemployment in the Czech Republic in the three selected years (2008, 2010 and 2014). It is based on the Czech Statistical Office data collected from the Labour Force Sample Survey. Lognormal probability distribution (unimodal positively skewed heavy-tailed distribution) is used in the Accelerated Failure Time regression model including the following factor explanatory variables: years (three levels, 2008, 2010, 2014), gender (two levels), education (four levels), five-year age groups (nine levels) and municipality size (five levels). Apart from this parametric model, Turnbull's nonparametric estimator of the survival function is evaluated for subsamples defined by the year, gender and education. The effects of education and gender -a strong positive one of the former and less significant one of the latter, respectively -are quantified and assessed.

No abstract

The aim of this paper is to examine the probability distribution of wealth of the richest persons in the world based on estimates from the CEOWORLD magazine's rich list for March 2019. Since one can safely assume that there are a tiny number of people out of the whole world's population in this list, we basically deal with the very right tail of the wealth distribution, which should according to the Pickands-Balkema-de Haan theorem follow a generalized Pareto distribution. We discuss in this paper not only different estimates of this distribution with an emphasis on the shape parameter per se but also their behavior throughout bootstrap samples. Among the main findings is the observation based on the maximum to sum plot and parametric estimates that there is high probability of infinite variance. This could have a serious impact on estimates of inequality measures. The whole distribution follows nearly a Pareto distribution, whereas the very right tail is closer to an exponential. The bootstrap technique confirms that maximum likelihood estimates are almost normally distributed, but they overestimate variance. Estimates via the method of L-moments diverge from the normal distribution. The correlation of the parameter estimates is moderately negative, which is demonstrated in a simultaneous confidence region.

The Use of Survival Analysis Methods for the Modelling of Unemployment in the Czech Republic Unemployment belongs to the most serious economic and social problems of developed countries. The problem can be described by the unemployment rate or number of the unemployed, in this text a duration of unemployment is of interest. The unemployment duration in the Czech Republic in 2008, 2010 and 2014 is analysed with the use of survival analysis methods and a finite mixture of lognormal distributions is used to describe an overall distribution of unemployment spell as well as the component distributions given by gender and education of the unemployed. Data from the Labour Force Sample Survey (performed by the Czech Statistical Office) are used for the statistical analysis. The unemployment duration is given in the questionnaire of the survey in intervals, we supposed the data to be right or interval censored, exact values of the unemployment duration are not included in the data. The strong positive effect of education on the duration of unemployment is quantified, as well as a less distinctive gender gap. An increase in unemployment duration is quantified for the period of economic crisis with respect to periods before (2008) and after (2014) the crisis.Keywords: unemployment duration, censored data, finite mixture of probability distributions, survival analysis JEL Classification: C41, J64, C24 ÚvodNezaměstnanost je problémem všech tržních ekonomik, proto je jejímu zkoumání věno-vána pozornost širokého spektra analytiků a výsledky týkající se nezaměstnanosti (napří-klad aktuální hodnoty míry nezaměstnanosti) jsou očekávány s velkým zájmem státních orgánů, ekonomických subjektů i laické veřejnosti. Obdobně velkým problémem, jako je počet nezaměstnaných, respektive míra nezaměstnanosti, je také doba, po kterou nezaměstnaní novou práci hledají. V případě nezaměstnanosti delší než jeden rok hovoříme o dlouhodobé nezaměstnanosti, míra dlouhodobé nezaměstnanosti je také sledována jako důležitá charakteristika stavu trhu práce a také ekonomiky jako celku (Krueger a kol., 2014). Počet nezaměstnaných a jejich strukturu, počet volných míst a délku nezaměstna-nosti ovlivňuje mnoho nejrůznějších faktorů. Vyhledávat tyto faktory, popsat a kvantifikovat jejich vliv, případně následně formulovat a realizovat opatření, která by se snažila

The aim of the paper is to compare three different approaches to modelling the distribution with supposed heavy tail using data about unemployment duration in the Czech republic -two models of spliced distributions and model with dynamic weights. All three aproaches are based on the Pickands-Balkema-de Haan theorem and incorporate the parametric distribution of the body of the probability distribution and generalized Pareto distribution used for the modelling of the right tail. The data come from the Labour Force Survey from three distinct periods -before the last economic crisis, during it and during the recovery from the crisis. The paper shows the differences between the right tails of the three periods and discussion of the used methods.